PyBullet Swarm Sim

10 Built-in Swarm Algorithms

PyBullet Swarm Sim includes a diverse library of foundational and advanced swarm intelligence algorithms. All algorithms implement the BaseAlgorithm interface and generate continuous velocity vectors based on the collective SwarmState.

• Reynolds Boids: The classic flocking model relying on Separation, Alignment, and Cohesion to simulate bird-like flight.
• Formation Flight: A structural planner that maintains precise geometric shapes like Lines, V-shapes, Grids, Rings, and Helices.
• Hover Swarm: A baseline fixed-point station-keeping algorithm to evaluate core physics and PID stability.
• PSO (Particle Swarm Optimization): A search algorithm where drones act as particles converging on a globally optimal target.
• ACO (Ant Colony Optimization): A path-planning strategy where agents lay digital pheromones to discover and reinforce optimal routes.
• Consensus: A distributed algorithm driving the swarm to rendezvous at a shared central point or reach state agreement.
• APF (Artificial Potential Fields): Uses attractive forces toward a goal and repulsive forces from obstacles and peers for dynamic navigation.
• ABC (Artificial Bee Colony): A foraging dynamics model splitting the swarm into Employed, Onlooker, and Scout roles for exploration.
• Voronoi Coverage: Utilizes Lloyd's algorithm to compute spatial Voronoi tessellations, optimally dispersing drones to maximize area coverage.
• MARL (PPO): A Multi-Agent Reinforcement Learning policy trained using parameter-sharing to achieve cooperative, collision-free goal seeking.

Custom Algorithm Upload (.py)

The dashboard provides a powerful feature to test your own experimental algorithms instantly without touching the core codebase or restarting the server. You can upload a custom Python (.py) file directly through the UI.

How it works

1. Create a Python file containing a class that inherits from BaseAlgorithm.
2. Implement the compute(self, state) method.
3. Upload the file in the "Upload Algorithm (.py)" section on the dashboard and click Run.

Your method will receive a SwarmState object every timestep containing the absolute positions and velocities of all drones, and it must return an (N, 3) NumPy array representing the target velocity vectors.

from swarm_sim.algorithms.base_algorithm import BaseAlgorithm
import numpy as np

class MyCustomSwarm(BaseAlgorithm):
    def __init__(self, num_drones, **kwargs):
        super().__init__(num_drones)
        self.speed = 1.0
        
    def compute(self, state):
        # state.positions: (N, 3) array of drone absolute (x,y,z) coordinates
        # state.velocities: (N, 3) array of drone velocities
        
        target_velocities = np.zeros((self.n, 3))
        # Example logic: Command all drones to move straight up
        target_velocities[:, 2] = self.speed 
        
        return target_velocities